Temporal median implementation accuracy

[thinkoid]

After reading the Calderara paper linked in the Temporal Median implementation (see section 2.2) I believe that the implementation is not faithful to the description of the algorithm. In particular the low and hight thresholds used to binarize the foreground mask are not per-pixel thresholds, but global.